add arize phoenix integration

backend/Makefile

@@ -1,7 +1,7 @@
 run:
 	echo "Running in local mode."
-	docker compose create db localstack
-	docker compose start db localstack
+	docker compose create db localstack phoenix
+	docker compose start db localstack phoenix
 	poetry run start
 
 run_docker:

backend/README.md

@@ -31,6 +31,11 @@ For any issues in setting up the above or during the rest of your development, y
 - [Open & already closed Github Issues](https://github.com/run-llama/sec-insights/issues?q=is%3Aissue+is%3Aclosed)
 - The [#sec-insights discord channel](https://discord.com/channels/1059199217496772688/1150942525968879636)
 
+## LLM Observability
+
+This project will automatically spin up a local version of [Arize Phoenix](https://phoenix.arize.com/) for you and send traces to it as you're using the chat interface.
+Arize Phoenix is a open source LLM observability & evaluation tool. LlamaIndex's event instrumentation system is deeply integrated with Arize Phoenix to make it easier for you to debug your LLM application during development. Simply open the Arize Phoenix Dashboard at [`http://localhost:6006/`](http://localhost:6006/) when running SEC Insights locally to see the traced calls to LLMs, Embedding Models, Vector DBs, and more.
+
 ## Scripts
 The `scripts/` folder contains several scripts that are useful for both operations and development.
 

backend/app/main.py

@@ -11,6 +11,7 @@ from alembic import script
 from alembic.runtime import migration
 from sqlalchemy.engine import create_engine, Engine
 from llama_index.core.node_parser.text.utils import split_by_sentence_tokenizer
+import llama_index.core
 
 from app.api.api import api_router
 from app.db.wait_for_db import check_database_connection
@@ -89,6 +90,10 @@ async def lifespan(app: FastAPI):
     except FileExistsError:
         # Sometimes seen in deployments, should be benign.
         logger.info("Tried to re-download NLTK files but already exists.")
+
+    if not settings.RENDER:
+        llama_index.core.set_global_handler("arize_phoenix")
+
     yield
     # This section is run on app shutdown
     await vector_store.close()

backend/docker-compose.yml

@@ -11,6 +11,7 @@ services:
       - "127.0.0.1:8000:8000"
     depends_on:
       - db
+      - phoenix
     env_file:
       - .env
       - .env.docker
@@ -41,5 +42,14 @@ services:
       - "${LOCALSTACK_VOLUME_DIR:-./volume}:/var/lib/localstack"
       - "/var/run/docker.sock:/var/run/docker.sock"
 
+  # useful for local workflow debugging
+  # taken from here: https://docs.arize.com/phoenix/deployment/docker#postgresql
+  phoenix:
+    image: arizephoenix/phoenix:latest # Must be greater than 4.0 version to work
+    ports:
+      - 6006:6006 # PHOENIX_PORT
+      - 4317:4317 # PHOENIX_GRPC_PORT
+      - 9090:9090 # [Optional] PROMETHEUS PORT IF ENABLED
+
 volumes:
   postgres_data:

backend/pyproject.toml

@@ -51,6 +51,8 @@ fire = "^0.5.0"
 sec-edgar-downloader = "~5.0"
 pytickersymbols = "^1.13.0"
 awscli-local = "^0.20"
+llama-index-callbacks-arize-phoenix = "^0.4.0"
+arize-phoenix = "^8.12.1"
 
 [tool.poetry.scripts]
 start = "app.main:start"